{
  "id": 8326,
  "title": "Continuous Probabilites better than discrete classes, is it acceptable?",
  "url": "/competitions/seizure-detection/discussion/8326",
  "author_name": "",
  "post_date": "2014-05-30T09:51:42.120Z",
  "votes": null,
  "comment_count": 3,
  "views": 1176,
  "content": "<p>Hi, I was wondering if it is ok to submit probablilits between 0 and 1 instead of discrete classification of 0 and 1 for early and seisures. I have noticed that AUC can be improved when submitting values between 0 and 1 instead of discrete 0 and 1. The submission system do accept continuous values such as:</p>\n<p>Dog_1_test_segment_1.mat,0.11,0.05<br>Dog_1_test_segment_2.mat,0.22,0.11<br>Dog_1_test_segment_3.mat,098,0.12</p>\n<p>Is this acceptable?</p>",
  "messages": [
    {
      "id": "47321",
      "postDate": "05/30/2014 09:51:42",
      "content": "<p>Hi, I was wondering if it is ok to submit probablilits between 0 and 1 instead of discrete classification of 0 and 1 for early and seisures. I have noticed that AUC can be improved when submitting values between 0 and 1 instead of discrete 0 and 1. The submission system do accept continuous values such as:</p>\n<p>Dog_1_test_segment_1.mat,0.11,0.05<br>Dog_1_test_segment_2.mat,0.22,0.11<br>Dog_1_test_segment_3.mat,098,0.12</p>\n<p>Is this acceptable?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "47322",
      "postDate": "05/30/2014 10:03:19",
      "content": "<p>You have to submit probabilities only.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "48523",
      "postDate": "06/02/2014 22:13:44",
      "content": "<p>&quot;Participants&nbsp;should submit answers in a CSV file with classifications for each class&quot;</p>\n<p>That is what evaluation page says, so what I deduce from that is that we have to threshold the probabilities to binary classification</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "48524",
      "postDate": "06/02/2014 22:16:44",
      "content": "<p>You may (and should, if you want to win) submit continuous probabilities as your predictions.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 47322,
      "author_name": "abhishek",
      "author_url": "",
      "post_date": "05/30/2014 10:03:19",
      "content": "<p>You have to submit probabilities only.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 48523,
      "author_name": "chaotic",
      "author_url": "",
      "post_date": "06/02/2014 22:13:44",
      "content": "<p>&quot;Participants&nbsp;should submit answers in a CSV file with classifications for each class&quot;</p>\n<p>That is what evaluation page says, so what I deduce from that is that we have to threshold the probabilities to binary classification</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 48524,
      "author_name": "wcukierski",
      "author_url": "",
      "post_date": "06/02/2014 22:16:44",
      "content": "<p>You may (and should, if you want to win) submit continuous probabilities as your predictions.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "47321": "",
    "47322": "",
    "48523": "",
    "48524": ""
  },
  "source": "meta"
}